Home | History | Annotate | Line # | Download | only in dev
vnd.c revision 1.40
      1 /*	$NetBSD: vnd.c,v 1.40 1997/06/08 15:59:41 pk Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1988 University of Utah.
      5  * Copyright (c) 1990, 1993
      6  *	The Regents of the University of California.  All rights reserved.
      7  *
      8  * This code is derived from software contributed to Berkeley by
      9  * the Systems Programming Group of the University of Utah Computer
     10  * Science Department.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the University of
     23  *	California, Berkeley and its contributors.
     24  * 4. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  * from: Utah $Hdr: vn.c 1.13 94/04/02$
     41  *
     42  *	@(#)vn.c	8.6 (Berkeley) 4/1/94
     43  */
     44 
     45 /*
     46  * Vnode disk driver.
     47  *
     48  * Block/character interface to a vnode.  Allows one to treat a file
     49  * as a disk (e.g. build a filesystem in it, mount it, etc.).
     50  *
     51  * NOTE 1: This uses the VOP_BMAP/VOP_STRATEGY interface to the vnode
     52  * instead of a simple VOP_RDWR.  We do this to avoid distorting the
     53  * local buffer cache.
     54  *
     55  * NOTE 2: There is a security issue involved with this driver.
     56  * Once mounted all access to the contents of the "mapped" file via
     57  * the special file is controlled by the permissions on the special
     58  * file, the protection of the mapped file is ignored (effectively,
     59  * by using root credentials in all transactions).
     60  *
     61  * NOTE 3: Doesn't interact with leases, should it?
     62  */
     63 
     64 #include <sys/param.h>
     65 #include <sys/systm.h>
     66 #include <sys/namei.h>
     67 #include <sys/proc.h>
     68 #include <sys/errno.h>
     69 #include <sys/buf.h>
     70 #include <sys/malloc.h>
     71 #include <sys/ioctl.h>
     72 #include <sys/disklabel.h>
     73 #include <sys/device.h>
     74 #include <sys/disk.h>
     75 #include <sys/stat.h>
     76 #include <sys/mount.h>
     77 #include <sys/vnode.h>
     78 #include <sys/file.h>
     79 #include <sys/uio.h>
     80 #include <sys/conf.h>
     81 
     82 #include <miscfs/specfs/specdev.h>
     83 
     84 #include <dev/vndioctl.h>
     85 
     86 #ifdef DEBUG
     87 int dovndcluster = 1;
     88 int vnddebug = 0x00;
     89 #define VDB_FOLLOW	0x01
     90 #define VDB_INIT	0x02
     91 #define VDB_IO		0x04
     92 #endif
     93 
     94 #define b_cylin	b_resid
     95 
     96 #define	vndunit(x)	DISKUNIT(x)
     97 
     98 struct vndxfer {
     99 	struct buf	*vx_bp;		/* Pointer to parent buffer */
    100 	int		vx_error;
    101 	int		vx_pending;	/* # of pending aux buffers */
    102 };
    103 
    104 struct vndbuf {
    105 	struct buf	vb_buf;
    106 	struct vndxfer	*vb_xfer;
    107 };
    108 
    109 #define	getvndxfer()	\
    110 	((struct vndxfer *)malloc(sizeof(struct vndxfer), M_DEVBUF, M_WAITOK))
    111 #define putvndxfer(vnx)	\
    112 	free((caddr_t)(vnx), M_DEVBUF)
    113 #define	getvndbuf()	\
    114 	((struct vndbuf *)malloc(sizeof(struct vndbuf), M_DEVBUF, M_WAITOK))
    115 #define putvndbuf(vbp)	\
    116 	free((caddr_t)(vbp), M_DEVBUF)
    117 
    118 struct vnd_softc {
    119 	int		 sc_flags;	/* flags */
    120 	size_t		 sc_size;	/* size of vnd */
    121 	struct vnode	*sc_vp;		/* vnode */
    122 	struct ucred	*sc_cred;	/* credentials */
    123 	int		 sc_maxactive;	/* max # of active requests */
    124 	struct buf	 sc_tab;	/* transfer queue */
    125 	char		 sc_xname[8];	/* XXX external name */
    126 	struct disk	 sc_dkdev;	/* generic disk device info */
    127 };
    128 
    129 /* sc_flags */
    130 #define	VNF_ALIVE	0x01
    131 #define VNF_INITED	0x02
    132 #define VNF_WANTED	0x40
    133 #define VNF_LOCKED	0x80
    134 
    135 struct vnd_softc *vnd_softc;
    136 int numvnd = 0;
    137 
    138 /* called by main() at boot time */
    139 void	vndattach __P((int));
    140 
    141 void	vndclear __P((struct vnd_softc *));
    142 void	vndstart __P((struct vnd_softc *));
    143 int	vndsetcred __P((struct vnd_softc *, struct ucred *));
    144 void	vndthrottle __P((struct vnd_softc *, struct vnode *));
    145 void	vndiodone __P((struct buf *));
    146 void	vndshutdown __P((void));
    147 
    148 static	int vndlock __P((struct vnd_softc *));
    149 static	void vndunlock __P((struct vnd_softc *));
    150 
    151 void
    152 vndattach(num)
    153 	int num;
    154 {
    155 	char *mem;
    156 	register u_long size;
    157 
    158 	if (num <= 0)
    159 		return;
    160 	size = num * sizeof(struct vnd_softc);
    161 	mem = malloc(size, M_DEVBUF, M_NOWAIT);
    162 	if (mem == NULL) {
    163 		printf("WARNING: no memory for vnode disks\n");
    164 		return;
    165 	}
    166 	bzero(mem, size);
    167 	vnd_softc = (struct vnd_softc *)mem;
    168 	numvnd = num;
    169 }
    170 
    171 int
    172 vndopen(dev, flags, mode, p)
    173 	dev_t dev;
    174 	int flags, mode;
    175 	struct proc *p;
    176 {
    177 	int unit = vndunit(dev);
    178 	struct vnd_softc *sc;
    179 	int error = 0, part, pmask;
    180 
    181 	/*
    182 	 * XXX Should support disklabels.
    183 	 */
    184 
    185 #ifdef DEBUG
    186 	if (vnddebug & VDB_FOLLOW)
    187 		printf("vndopen(%x, %x, %x, %p)\n", dev, flags, mode, p);
    188 #endif
    189 	if (unit >= numvnd)
    190 		return (ENXIO);
    191 	sc = &vnd_softc[unit];
    192 
    193 	if ((error = vndlock(sc)) != 0)
    194 		return (error);
    195 
    196 	part = DISKPART(dev);
    197 	pmask = (1 << part);
    198 
    199 	/* Prevent our unit from being unconfigured while open. */
    200 	switch (mode) {
    201 	case S_IFCHR:
    202 		sc->sc_dkdev.dk_copenmask |= pmask;
    203 		break;
    204 
    205 	case S_IFBLK:
    206 		sc->sc_dkdev.dk_bopenmask |= pmask;
    207 		break;
    208 	}
    209 	sc->sc_dkdev.dk_openmask =
    210 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    211 
    212 	vndunlock(sc);
    213 	return (0);
    214 }
    215 
    216 int
    217 vndclose(dev, flags, mode, p)
    218 	dev_t dev;
    219 	int flags, mode;
    220 	struct proc *p;
    221 {
    222 	int unit = vndunit(dev);
    223 	struct vnd_softc *sc;
    224 	int error = 0, part;
    225 
    226 #ifdef DEBUG
    227 	if (vnddebug & VDB_FOLLOW)
    228 		printf("vndclose(%x, %x, %x, %p)\n", dev, flags, mode, p);
    229 #endif
    230 
    231 	if (unit >= numvnd)
    232 		return (ENXIO);
    233 	sc = &vnd_softc[unit];
    234 
    235 	if ((error = vndlock(sc)) != 0)
    236 		return (error);
    237 
    238 	part = DISKPART(dev);
    239 
    240 	/* ...that much closer to allowing unconfiguration... */
    241 	switch (mode) {
    242 	case S_IFCHR:
    243 		sc->sc_dkdev.dk_copenmask &= ~(1 << part);
    244 		break;
    245 
    246 	case S_IFBLK:
    247 		sc->sc_dkdev.dk_bopenmask &= ~(1 << part);
    248 		break;
    249 	}
    250 	sc->sc_dkdev.dk_openmask =
    251 	    sc->sc_dkdev.dk_copenmask | sc->sc_dkdev.dk_bopenmask;
    252 
    253 	vndunlock(sc);
    254 	return (0);
    255 }
    256 
    257 /*
    258  * Break the request into bsize pieces and submit using VOP_BMAP/VOP_STRATEGY.
    259  * Note that this driver can only be used for swapping over NFS on the hp
    260  * since nfs_strategy on the vax cannot handle u-areas and page tables.
    261  */
    262 void
    263 vndstrategy(bp)
    264 	register struct buf *bp;
    265 {
    266 	int unit = vndunit(bp->b_dev);
    267 	struct vnd_softc *vnd = &vnd_softc[unit];
    268 	struct vndbuf *nbp;
    269 	struct vndxfer *vnx;
    270 	int bn, bsize, resid;
    271 	caddr_t addr;
    272 	int sz, flags, error;
    273 
    274 #ifdef DEBUG
    275 	if (vnddebug & VDB_FOLLOW)
    276 		printf("vndstrategy(%p): unit %d\n", bp, unit);
    277 #endif
    278 	if ((vnd->sc_flags & VNF_INITED) == 0) {
    279 		bp->b_error = ENXIO;
    280 		bp->b_flags |= B_ERROR;
    281 		biodone(bp);
    282 		return;
    283 	}
    284 	bn = bp->b_blkno;
    285 	sz = howmany(bp->b_bcount, DEV_BSIZE);
    286 	bp->b_resid = bp->b_bcount;
    287 	if (bn < 0 || bn + sz > vnd->sc_size) {
    288 		if (bn != vnd->sc_size) {
    289 			bp->b_error = EINVAL;
    290 			bp->b_flags |= B_ERROR;
    291 		}
    292 		biodone(bp);
    293 		return;
    294 	}
    295 	bn = dbtob(bn);
    296  	bsize = vnd->sc_vp->v_mount->mnt_stat.f_iosize;
    297 	addr = bp->b_data;
    298 	flags = bp->b_flags | B_CALL;
    299 
    300 	/* Allocate a header for this transfer and link it to the buffer */
    301 	vnx = getvndxfer();
    302 	vnx->vx_error = 0;
    303 	vnx->vx_pending = 0;
    304 	vnx->vx_bp = bp;
    305 
    306 	for (resid = bp->b_resid; resid; resid -= sz) {
    307 		struct vnode *vp;
    308 		daddr_t nbn;
    309 		int off, s, nra;
    310 
    311 		nra = 0;
    312 		VOP_LOCK(vnd->sc_vp);
    313 		error = VOP_BMAP(vnd->sc_vp, bn / bsize, &vp, &nbn, &nra);
    314 		VOP_UNLOCK(vnd->sc_vp);
    315 
    316 		if (error == 0 && (long)nbn == -1)
    317 			error = EIO;
    318 
    319 		/*
    320 		 * If there was an error or a hole in the file...punt.
    321 		 * Note that we may have to wait for any operations
    322 		 * that we have already fired off before releasing
    323 		 * the buffer.
    324 		 *
    325 		 * XXX we could deal with holes here but it would be
    326 		 * a hassle (in the write case).
    327 		 */
    328 		if (error) {
    329 			vnx->vx_error = error;
    330 			s = splbio();
    331 			if (vnx->vx_pending == 0) {
    332 				bp->b_error = error;
    333 				bp->b_flags |= B_ERROR;
    334 				putvndxfer(vnx);
    335 				biodone(bp);
    336 			}
    337 			splx(s);
    338 			return;
    339 		}
    340 
    341 #ifdef DEBUG
    342 		if (!dovndcluster)
    343 			nra = 0;
    344 #endif
    345 
    346 		if ((off = bn % bsize) != 0)
    347 			sz = bsize - off;
    348 		else
    349 			sz = (1 + nra) * bsize;
    350 		if (resid < sz)
    351 			sz = resid;
    352 #ifdef DEBUG
    353 		if (vnddebug & VDB_IO)
    354 			printf("vndstrategy: vp %p/%p bn %x/%x sz %x\n",
    355 			    vnd->sc_vp, vp, bn, nbn, sz);
    356 #endif
    357 
    358 		nbp = getvndbuf();
    359 		nbp->vb_buf.b_flags = flags;
    360 		nbp->vb_buf.b_bcount = sz;
    361 		nbp->vb_buf.b_bufsize = bp->b_bufsize;
    362 		nbp->vb_buf.b_error = 0;
    363 		if (vp->v_type == VBLK || vp->v_type == VCHR)
    364 			nbp->vb_buf.b_dev = vp->v_rdev;
    365 		else
    366 			nbp->vb_buf.b_dev = NODEV;
    367 		nbp->vb_buf.b_data = addr;
    368 		nbp->vb_buf.b_blkno = nbn + btodb(off);
    369 		nbp->vb_buf.b_proc = bp->b_proc;
    370 		nbp->vb_buf.b_iodone = vndiodone;
    371 		nbp->vb_buf.b_vp = vp;
    372 		nbp->vb_buf.b_rcred = vnd->sc_cred;	/* XXX crdup? */
    373 		nbp->vb_buf.b_wcred = vnd->sc_cred;	/* XXX crdup? */
    374 		if (bp->b_dirtyend == 0) {
    375 			nbp->vb_buf.b_dirtyoff = 0;
    376 			nbp->vb_buf.b_dirtyend = sz;
    377 		} else {
    378 			nbp->vb_buf.b_dirtyoff =
    379 			    max(0, bp->b_dirtyoff - (bp->b_bcount - resid));
    380 			nbp->vb_buf.b_dirtyend =
    381 			    min(sz,
    382 				max(0, bp->b_dirtyend - (bp->b_bcount-resid)));
    383 		}
    384 		if (bp->b_validend == 0) {
    385 			nbp->vb_buf.b_validoff = 0;
    386 			nbp->vb_buf.b_validend = sz;
    387 		} else {
    388 			nbp->vb_buf.b_validoff =
    389 			    max(0, bp->b_validoff - (bp->b_bcount - resid));
    390 			nbp->vb_buf.b_validend =
    391 			    min(sz,
    392 				max(0, bp->b_validend - (bp->b_bcount-resid)));
    393 		}
    394 
    395 		nbp->vb_xfer = vnx;
    396 
    397 		/*
    398 		 * Just sort by block number
    399 		 */
    400 		nbp->vb_buf.b_cylin = nbp->vb_buf.b_blkno;
    401 		s = splbio();
    402 		vnx->vx_pending++;
    403 		disksort(&vnd->sc_tab, &nbp->vb_buf);
    404 		if (vnd->sc_tab.b_active < vnd->sc_maxactive) {
    405 			vnd->sc_tab.b_active++;
    406 			vndstart(vnd);
    407 		}
    408 		splx(s);
    409 		bn += sz;
    410 		addr += sz;
    411 	}
    412 }
    413 
    414 /*
    415  * Feed requests sequentially.
    416  * We do it this way to keep from flooding NFS servers if we are connected
    417  * to an NFS file.  This places the burden on the client rather than the
    418  * server.
    419  */
    420 void
    421 vndstart(vnd)
    422 	register struct vnd_softc *vnd;
    423 {
    424 	register struct buf *bp;
    425 
    426 	/*
    427 	 * Dequeue now since lower level strategy routine might
    428 	 * queue using same links
    429 	 */
    430 	bp = vnd->sc_tab.b_actf;
    431 	vnd->sc_tab.b_actf = bp->b_actf;
    432 #ifdef DEBUG
    433 	if (vnddebug & VDB_IO)
    434 		printf("vndstart(%ld): bp %p vp %p blkno %x addr %p cnt %lx\n",
    435 		    (long) (vnd-vnd_softc), bp, bp->b_vp, bp->b_blkno,
    436 		    bp->b_data, bp->b_bcount);
    437 #endif
    438 
    439 	/* Instrumentation. */
    440 	disk_busy(&vnd->sc_dkdev);
    441 
    442 	if ((bp->b_flags & B_READ) == 0)
    443 		bp->b_vp->v_numoutput++;
    444 	VOP_STRATEGY(bp);
    445 }
    446 
    447 void
    448 vndiodone(bp)
    449 	struct buf *bp;
    450 {
    451 	register struct vndbuf *vbp = (struct vndbuf *) bp;
    452 	register struct vndxfer *vnx = (struct vndxfer *)vbp->vb_xfer;
    453 	register struct buf *pbp = vnx->vx_bp;
    454 	register struct vnd_softc *vnd = &vnd_softc[vndunit(pbp->b_dev)];
    455 	int s, resid;
    456 
    457 	s = splbio();
    458 #ifdef DEBUG
    459 	if (vnddebug & VDB_IO)
    460 		printf("vndiodone(%ld): vbp %p vp %p blkno %x addr %p cnt %lx\n",
    461 		    (long) (vnd-vnd_softc), vbp, vbp->vb_buf.b_vp,
    462 		    vbp->vb_buf.b_blkno, vbp->vb_buf.b_data,
    463 		    vbp->vb_buf.b_bcount);
    464 #endif
    465 
    466 	resid = vbp->vb_buf.b_bcount - vbp->vb_buf.b_resid;
    467 	pbp->b_resid -= resid;
    468 	disk_unbusy(&vnd->sc_dkdev, resid);
    469 	vnx->vx_pending--;
    470 
    471 	if (vbp->vb_buf.b_error) {
    472 #ifdef DEBUG
    473 		if (vnddebug & VDB_IO)
    474 			printf("vndiodone: vbp %p error %d\n", vbp,
    475 			    vbp->vb_buf.b_error);
    476 #endif
    477 		vnx->vx_error = vbp->vb_buf.b_error;
    478 	}
    479 	putvndbuf(vbp);
    480 
    481 	/*
    482 	 * Wrap up this transaction if it has run to completion or, in
    483 	 * case of an error, when all auxiliary buffers have returned.
    484 	 */
    485 	if (pbp->b_resid == 0 || (vnx->vx_error && vnx->vx_pending == 0)) {
    486 
    487 		if (vnx->vx_error != 0) {
    488 			pbp->b_flags |= B_ERROR;
    489 			pbp->b_error = vnx->vx_error;
    490 		}
    491 		putvndxfer(vnx);
    492 #ifdef DEBUG
    493 		if (vnddebug & VDB_IO)
    494 			printf("vndiodone: pbp %p iodone\n", pbp);
    495 #endif
    496 		biodone(pbp);
    497 	}
    498 
    499 	if (vnd->sc_tab.b_actf)
    500 		vndstart(vnd);
    501 	else
    502 		vnd->sc_tab.b_active--;
    503 	splx(s);
    504 }
    505 
    506 /* ARGSUSED */
    507 int
    508 vndread(dev, uio, flags)
    509 	dev_t dev;
    510 	struct uio *uio;
    511 	int flags;
    512 {
    513 	int unit = vndunit(dev);
    514 	struct vnd_softc *sc;
    515 
    516 #ifdef DEBUG
    517 	if (vnddebug & VDB_FOLLOW)
    518 		printf("vndread(%x, %p)\n", dev, uio);
    519 #endif
    520 
    521 	if (unit >= numvnd)
    522 		return (ENXIO);
    523 	sc = &vnd_softc[unit];
    524 
    525 	if ((sc->sc_flags & VNF_INITED) == 0)
    526 		return (ENXIO);
    527 
    528 	return (physio(vndstrategy, NULL, dev, B_READ, minphys, uio));
    529 }
    530 
    531 /* ARGSUSED */
    532 int
    533 vndwrite(dev, uio, flags)
    534 	dev_t dev;
    535 	struct uio *uio;
    536 	int flags;
    537 {
    538 	int unit = vndunit(dev);
    539 	struct vnd_softc *sc;
    540 
    541 #ifdef DEBUG
    542 	if (vnddebug & VDB_FOLLOW)
    543 		printf("vndwrite(%x, %p)\n", dev, uio);
    544 #endif
    545 
    546 	if (unit >= numvnd)
    547 		return (ENXIO);
    548 	sc = &vnd_softc[unit];
    549 
    550 	if ((sc->sc_flags & VNF_INITED) == 0)
    551 		return (ENXIO);
    552 
    553 	return (physio(vndstrategy, NULL, dev, B_WRITE, minphys, uio));
    554 }
    555 
    556 /* ARGSUSED */
    557 int
    558 vndioctl(dev, cmd, data, flag, p)
    559 	dev_t dev;
    560 	u_long cmd;
    561 	caddr_t data;
    562 	int flag;
    563 	struct proc *p;
    564 {
    565 	int unit = vndunit(dev);
    566 	register struct vnd_softc *vnd;
    567 	struct vnd_ioctl *vio;
    568 	struct vattr vattr;
    569 	struct nameidata nd;
    570 	int error, part, pmask;
    571 
    572 #ifdef DEBUG
    573 	if (vnddebug & VDB_FOLLOW)
    574 		printf("vndioctl(%x, %lx, %p, %x, %p): unit %d\n",
    575 		    dev, cmd, data, flag, p, unit);
    576 #endif
    577 	error = suser(p->p_ucred, &p->p_acflag);
    578 	if (error)
    579 		return (error);
    580 	if (unit >= numvnd)
    581 		return (ENXIO);
    582 
    583 	vnd = &vnd_softc[unit];
    584 	vio = (struct vnd_ioctl *)data;
    585 	switch (cmd) {
    586 
    587 	case VNDIOCSET:
    588 		if (vnd->sc_flags & VNF_INITED)
    589 			return (EBUSY);
    590 
    591 		if ((error = vndlock(vnd)) != 0)
    592 			return (error);
    593 
    594 		/*
    595 		 * Always open for read and write.
    596 		 * This is probably bogus, but it lets vn_open()
    597 		 * weed out directories, sockets, etc. so we don't
    598 		 * have to worry about them.
    599 		 */
    600 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, vio->vnd_file, p);
    601 		if ((error = vn_open(&nd, FREAD|FWRITE, 0)) != 0) {
    602 			vndunlock(vnd);
    603 			return(error);
    604 		}
    605 		error = VOP_GETATTR(nd.ni_vp, &vattr, p->p_ucred, p);
    606 		if (error) {
    607 			VOP_UNLOCK(nd.ni_vp);
    608 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    609 			vndunlock(vnd);
    610 			return(error);
    611 		}
    612 		VOP_UNLOCK(nd.ni_vp);
    613 		vnd->sc_vp = nd.ni_vp;
    614 		vnd->sc_size = btodb(vattr.va_size);	/* note truncation */
    615 		if ((error = vndsetcred(vnd, p->p_ucred)) != 0) {
    616 			(void) vn_close(nd.ni_vp, FREAD|FWRITE, p->p_ucred, p);
    617 			vndunlock(vnd);
    618 			return(error);
    619 		}
    620 		vndthrottle(vnd, vnd->sc_vp);
    621 		vio->vnd_size = dbtob(vnd->sc_size);
    622 		vnd->sc_flags |= VNF_INITED;
    623 #ifdef DEBUG
    624 		if (vnddebug & VDB_INIT)
    625 			printf("vndioctl: SET vp %p size %lx\n",
    626 			    vnd->sc_vp, (unsigned long) vnd->sc_size);
    627 #endif
    628 
    629 		/* Attach the disk. */
    630 		bzero(vnd->sc_xname, sizeof(vnd->sc_xname));	/* XXX */
    631 		sprintf(vnd->sc_xname, "vnd%d", unit);		/* XXX */
    632 		vnd->sc_dkdev.dk_name = vnd->sc_xname;
    633 		disk_attach(&vnd->sc_dkdev);
    634 
    635 		vndunlock(vnd);
    636 
    637 		break;
    638 
    639 	case VNDIOCCLR:
    640 		if ((vnd->sc_flags & VNF_INITED) == 0)
    641 			return (ENXIO);
    642 
    643 		if ((error = vndlock(vnd)) != 0)
    644 			return (error);
    645 
    646 		/*
    647 		 * Don't unconfigure if any other partitions are open
    648 		 * or if both the character and block flavors of this
    649 		 * partition are open.
    650 		 */
    651 		part = DISKPART(dev);
    652 		pmask = (1 << part);
    653 		if ((vnd->sc_dkdev.dk_openmask & ~pmask) ||
    654 		    ((vnd->sc_dkdev.dk_bopenmask & pmask) &&
    655 		    (vnd->sc_dkdev.dk_copenmask & pmask))) {
    656 			vndunlock(vnd);
    657 			return (EBUSY);
    658 		}
    659 
    660 		vndclear(vnd);
    661 #ifdef DEBUG
    662 		if (vnddebug & VDB_INIT)
    663 			printf("vndioctl: CLRed\n");
    664 #endif
    665 
    666 		/* Detatch the disk. */
    667 		disk_detach(&vnd->sc_dkdev);
    668 
    669 		vndunlock(vnd);
    670 
    671 		break;
    672 
    673 	/*
    674 	 * XXX Should support disklabels.
    675 	 */
    676 
    677 	default:
    678 		return(ENOTTY);
    679 	}
    680 
    681 	return (0);
    682 }
    683 
    684 /*
    685  * Duplicate the current processes' credentials.  Since we are called only
    686  * as the result of a SET ioctl and only root can do that, any future access
    687  * to this "disk" is essentially as root.  Note that credentials may change
    688  * if some other uid can write directly to the mapped file (NFS).
    689  */
    690 int
    691 vndsetcred(vnd, cred)
    692 	register struct vnd_softc *vnd;
    693 	struct ucred *cred;
    694 {
    695 	struct uio auio;
    696 	struct iovec aiov;
    697 	char *tmpbuf;
    698 	int error;
    699 
    700 	vnd->sc_cred = crdup(cred);
    701 	tmpbuf = malloc(DEV_BSIZE, M_TEMP, M_WAITOK);
    702 
    703 	/* XXX: Horrible kludge to establish credentials for NFS */
    704 	aiov.iov_base = tmpbuf;
    705 	aiov.iov_len = min(DEV_BSIZE, dbtob(vnd->sc_size));
    706 	auio.uio_iov = &aiov;
    707 	auio.uio_iovcnt = 1;
    708 	auio.uio_offset = 0;
    709 	auio.uio_rw = UIO_READ;
    710 	auio.uio_segflg = UIO_SYSSPACE;
    711 	auio.uio_resid = aiov.iov_len;
    712 	VOP_LOCK(vnd->sc_vp);
    713 	error = VOP_READ(vnd->sc_vp, &auio, 0, vnd->sc_cred);
    714 	VOP_UNLOCK(vnd->sc_vp);
    715 
    716 	free(tmpbuf, M_TEMP);
    717 	return (error);
    718 }
    719 
    720 /*
    721  * Set maxactive based on FS type
    722  */
    723 void
    724 vndthrottle(vnd, vp)
    725 	register struct vnd_softc *vnd;
    726 	struct vnode *vp;
    727 {
    728 #ifdef NFS
    729 	extern int (**nfsv2_vnodeop_p) __P((void *));
    730 
    731 	if (vp->v_op == nfsv2_vnodeop_p)
    732 		vnd->sc_maxactive = 2;
    733 	else
    734 #endif
    735 		vnd->sc_maxactive = 8;
    736 
    737 	if (vnd->sc_maxactive < 1)
    738 		vnd->sc_maxactive = 1;
    739 }
    740 
    741 void
    742 vndshutdown()
    743 {
    744 	register struct vnd_softc *vnd;
    745 
    746 	for (vnd = &vnd_softc[0]; vnd < &vnd_softc[numvnd]; vnd++)
    747 		if (vnd->sc_flags & VNF_INITED)
    748 			vndclear(vnd);
    749 }
    750 
    751 void
    752 vndclear(vnd)
    753 	register struct vnd_softc *vnd;
    754 {
    755 	register struct vnode *vp = vnd->sc_vp;
    756 	struct proc *p = curproc;		/* XXX */
    757 
    758 #ifdef DEBUG
    759 	if (vnddebug & VDB_FOLLOW)
    760 		printf("vndclear(%p): vp %p\n", vnd, vp);
    761 #endif
    762 	vnd->sc_flags &= ~VNF_INITED;
    763 	if (vp == (struct vnode *)0)
    764 		panic("vndioctl: null vp");
    765 	(void) vn_close(vp, FREAD|FWRITE, vnd->sc_cred, p);
    766 	crfree(vnd->sc_cred);
    767 	vnd->sc_vp = (struct vnode *)0;
    768 	vnd->sc_cred = (struct ucred *)0;
    769 	vnd->sc_size = 0;
    770 }
    771 
    772 int
    773 vndsize(dev)
    774 	dev_t dev;
    775 {
    776 	int unit = vndunit(dev);
    777 	register struct vnd_softc *vnd = &vnd_softc[unit];
    778 
    779 	if (unit >= numvnd || (vnd->sc_flags & VNF_INITED) == 0)
    780 		return(-1);
    781 	return(vnd->sc_size);
    782 }
    783 
    784 int
    785 vnddump(dev, blkno, va, size)
    786 	dev_t dev;
    787 	daddr_t blkno;
    788 	caddr_t va;
    789 	size_t size;
    790 {
    791 
    792 	/* Not implemented. */
    793 	return ENXIO;
    794 }
    795 
    796 /*
    797  * Wait interruptibly for an exclusive lock.
    798  *
    799  * XXX
    800  * Several drivers do this; it should be abstracted and made MP-safe.
    801  */
    802 static int
    803 vndlock(sc)
    804 	struct vnd_softc *sc;
    805 {
    806 	int error;
    807 
    808 	while ((sc->sc_flags & VNF_LOCKED) != 0) {
    809 		sc->sc_flags |= VNF_WANTED;
    810 		if ((error = tsleep(sc, PRIBIO | PCATCH, "vndlck", 0)) != 0)
    811 			return (error);
    812 	}
    813 	sc->sc_flags |= VNF_LOCKED;
    814 	return (0);
    815 }
    816 
    817 /*
    818  * Unlock and wake up any waiters.
    819  */
    820 static void
    821 vndunlock(sc)
    822 	struct vnd_softc *sc;
    823 {
    824 
    825 	sc->sc_flags &= ~VNF_LOCKED;
    826 	if ((sc->sc_flags & VNF_WANTED) != 0) {
    827 		sc->sc_flags &= ~VNF_WANTED;
    828 		wakeup(sc);
    829 	}
    830 }
    831